Upload folder using huggingface_hub

llamagen-siglip-sb-block-causal/scripts/autoregressive/extract_codes_c2i.sh

@@ -0,0 +1,15 @@
+# !/bin/bash
+set -x
+
+torchrun \
+    --nnodes=1 --nproc_per_node=8 --node_rank=0  \
+    --master_port=12335 \
+    autoregressive/train/extract_codes_c2i.py \
+    --yml-path /home/ma-user/work/p84402465/llamagen-finetune-main/config/selftok_sd3_E31-512_modified.yml \
+    --vq-ckpt /home/ma-user/work/w84402926/SelfTok/llamagen/llamagen-finetune-main/weight/iter_237999.pth \
+    --sd3-pretrained /home/ma-user/work/w84402926/SelfTok/llamagen/llamagen-finetune-main/weight/sd3_medium.ckpt \
+    --data-path /home/ma-user/work/p84402465/imagenet \
+    --code-path /cache/code-imagenet-E31 \
+    --image-size 512
+
+

llamagen-siglip-sb-block-causal/scripts/autoregressive/extract_codes_t2i.sh

@@ -0,0 +1,15 @@
+# !/bin/bash
+set -x
+
+torchrun \
+    --nnodes=1 --nproc_per_node=8 --node_rank=0  \
+    --master_port=12335 \
+    autoregressive/train/extract_codes_t2i.py \
+    --yml-path /home/ma-user/work/p84402465/llamagen-finetune-main/config/selftok_sd3_E31-512_modified.yml \
+    --vq-ckpt /home/ma-user/work/w84402926/SelfTok/llamagen/llamagen-finetune-main/weight/iter_237999.pth \
+    --sd3-pretrained /home/ma-user/work/w84402926/SelfTok/llamagen/llamagen-finetune-main/weight/sd3_medium.ckpt \
+    --data-path /cache/imagenet \
+    --code-path /cache/code-imagenet-E31_cap \
+    --image-size 512
+
+

llamagen-siglip-sb-block-causal/scripts/autoregressive/sample_c2i.sh

@@ -0,0 +1,31 @@
+# # !/bin/bash
+# set -x
+
+# torchrun \
+# --nnodes=1 --nproc_per_node=8 --node_rank=0 \
+# --master_port=12345 \
+# autoregressive/sample/sample_c2i_ddp.py \
+# --vq-ckpt ./pretrained_models/vq_ds16_c2i.pt \
+# "$@"
+
+# # !/bin/bash
+# set -x
+
+torchrun \
+    --nnodes=1 --nproc_per_node=1 --node_rank=0 \
+    --master_port=12345 \
+    autoregressive/sample/sample_c2i.py \
+    --yml-path="/home/ma-user/work/p84402465/SelftokPipeline-master/configs/renderer/selftok_sd3_E31-512_renderer.yml" \
+    --vq-ckpt="/home/ma-user/work/p84402465/cache/models/pretrain_tokenizer_iter_331999.pth" \
+    --save-path="./sample_c2i_240k_cfg10_largebs.png" \
+    --gpt-model="GPT-XL" \
+    --gpt-ckpt="/home/ma-user/work/p84402465/cache/llamagen-E31-0811/checkpoint-30000/consolidated.pth" \
+    --sd3-pretrained="/home/ma-user/work/w84402926/SelfTok/llamagen/llamagen-finetune-main/weight/sd3_medium.ckpt" \
+    --gpt-type="c2i" \
+    --codebook-size=16384 \
+    --image-size=512 \
+    --cfg-scale 10 \
+	--top-p 1.0 \
+	--top-k 0 \
+	--temperature 1.0 \
+    --from-fsdp

llamagen-siglip-sb-block-causal/scripts/autoregressive/sample_t2i.sh

@@ -0,0 +1,54 @@
+# !/bin/bash
+set -x
+
+GPUS=1
+
+CFG=3.0
+
+MODEL_S3_PATH="s3://bucket-5125-guiyang/selftok/outputs/llamagen_selftok/Selftok-Llamagen-t2i-vit-sb-v16-blockcausal/checkpoint-505000"
+MOX_OVERWRITE=False
+
+parent_dir=$(basename "$(dirname "$MODEL_S3_PATH")")
+base_dir=$(basename "$MODEL_S3_PATH")
+local_model_path="/cache/ckpts/${parent_dir}/${base_dir}"
+
+echo "[INFO] model parent_dir = $parent_dir"
+echo "[INFO] model base_dir   = $base_dir"
+echo "[INFO] local_model_path = $local_model_path"
+
+python3 - <<EOF
+import os
+import moxing as mox
+
+model_s3 = "$MODEL_S3_PATH"
+local_path = "$local_model_path"
+overwrite = $MOX_OVERWRITE
+
+if not os.path.exists(local_path) or overwrite:
+    os.makedirs(os.path.dirname(local_path), exist_ok=True)
+    print(f"[COPY] {model_s3} -> {local_path}")
+    mox.file.copy_parallel(model_s3, local_path)
+    print("[COPY DONE]")
+else:
+    print(f"[SKIP] local model already exists: {local_path}")
+EOF
+
+torchrun \
+    --nnodes=1 --nproc_per_node=$GPUS --node_rank=0 \
+    --master_port=12346 \
+    autoregressive/sample/sample_t2i.py \
+    --yml-path="/home/ma-user/work/z84399568/selftok_sd3_vit_spatial_block_512res_eval_fsq/configs/mimo/selftok/spatial-block/v16.yml" \
+    --vq-ckpt="/home/ma-user/work/z84399568/selftok_sd3_vit_spatial_block_512res_eval_fsq/iter_155999.pth" \
+    --gpt-model="GPT-XL" \
+    --gpt-ckpt=$local_model_path \
+    --save-image="./results/imagenet_siglip2/${parent_dir}-${base_dir}-cfg-${CFG}" \
+    --sd3-pretrained="/home/ma-user/work/selftok/models/sd3_medium.ckpt" \
+    --t5-path="/home/ma-user/work/selftok/models" \
+    --gpt-type="t2i" \
+    --codebook-size=19683 \
+    --image-size=512 \
+    --cfg-scale $CFG \
+	--top-p 1.0 \
+	--top-k 0 \
+	--temperature 1.0 \
+

llamagen-siglip-sb-block-causal/scripts/autoregressive/sample_t2i_coco.sh

@@ -0,0 +1,11 @@
+# !/bin/bash
+set -x
+
+torchrun \
+--nnodes=1 --nproc_per_node=8 --node_rank=0 \
+--master_port=12346 \
+autoregressive/sample/sample_t2i_ddp.py \
+--prompt-csv evaluations/t2i/coco_captions.csv \
+--sample-dir samples_coco \
+--vq-ckpt ./pretrained_models/vq_ds16_t2i.pt \
+"$@"

llamagen-siglip-sb-block-causal/scripts/autoregressive/sample_t2i_ddp.sh

@@ -0,0 +1,54 @@
+# !/bin/bash
+set -x
+
+GPUS=8
+
+CFG=3.0
+
+MODEL_S3_PATH="s3://bucket-5125-guiyang/selftok/outputs/llamagen_selftok/Selftok-Llamagen-t2i-vit-sb-v16-blockcausal/checkpoint-485000"
+MOX_OVERWRITE=False
+
+parent_dir=$(basename "$(dirname "$MODEL_S3_PATH")")
+base_dir=$(basename "$MODEL_S3_PATH")
+local_model_path="/cache/ckpts/${parent_dir}/${base_dir}"
+
+echo "[INFO] model parent_dir = $parent_dir"
+echo "[INFO] model base_dir   = $base_dir"
+echo "[INFO] local_model_path = $local_model_path"
+
+python3 - <<EOF
+import os
+import moxing as mox
+
+model_s3 = "$MODEL_S3_PATH"
+local_path = "$local_model_path"
+overwrite = $MOX_OVERWRITE
+
+if not os.path.exists(local_path) or overwrite:
+    os.makedirs(os.path.dirname(local_path), exist_ok=True)
+    print(f"[COPY] {model_s3} -> {local_path}")
+    mox.file.copy_parallel(model_s3, local_path)
+    print("[COPY DONE]")
+else:
+    print(f"[SKIP] local model already exists: {local_path}")
+EOF
+
+torchrun \
+    --nnodes=1 --nproc_per_node=$GPUS --node_rank=0 \
+    --master_port=12346 \
+    autoregressive/sample/sample_t2i_ddp.py \
+    --yml-path="/home/ma-user/work/z84399568/selftok_sd3_vit_spatial_block_512res_eval_fsq/configs/mimo/selftok/spatial-block/v16.yml" \
+    --vq-ckpt="/home/ma-user/work/z84399568/selftok_sd3_vit_spatial_block_512res_eval_fsq/iter_155999.pth" \
+    --gpt-model="GPT-XL" \
+    --prompt-csv="/home/ma-user/work/p84402465/imagenet_process/imagenet_val_captions.csv" \
+    --gpt-ckpt=$local_model_path \
+    --sd3-pretrained="/home/ma-user/work/selftok/models/sd3_medium.ckpt" \
+    --t5-path="/home/ma-user/work/selftok/models" \
+    --gpt-type="t2i" \
+    --codebook-size=19683 \
+    --image-size=512 \
+    --cfg-scale $CFG \
+	--top-p 1.0 \
+	--top-k 0 \
+	--temperature 1.0 \
+

llamagen-siglip-sb-block-causal/scripts/autoregressive/sample_t2i_noreverse.sh

@@ -0,0 +1,54 @@
+# !/bin/bash
+set -x
+
+GPUS=1
+
+CFG=3.0
+
+MODEL_S3_PATH="s3://bucket-5125-guiyang/selftok/outputs/llamagen_selftok/Selftok-Llamagen-t2i-SigLip-noreverse/checkpoint-330000"
+MOX_OVERWRITE=False
+
+parent_dir=$(basename "$(dirname "$MODEL_S3_PATH")")
+base_dir=$(basename "$MODEL_S3_PATH")
+local_model_path="/cache/ckpts/${parent_dir}/${base_dir}"
+
+echo "[INFO] model parent_dir = $parent_dir"
+echo "[INFO] model base_dir   = $base_dir"
+echo "[INFO] local_model_path = $local_model_path"
+
+python3 - <<EOF
+import os
+import moxing as mox
+
+model_s3 = "$MODEL_S3_PATH"
+local_path = "$local_model_path"
+overwrite = $MOX_OVERWRITE
+
+if not os.path.exists(local_path) or overwrite:
+    os.makedirs(os.path.dirname(local_path), exist_ok=True)
+    print(f"[COPY] {model_s3} -> {local_path}")
+    mox.file.copy_parallel(model_s3, local_path)
+    print("[COPY DONE]")
+else:
+    print(f"[SKIP] local model already exists: {local_path}")
+EOF
+
+torchrun \
+    --nnodes=1 --nproc_per_node=$GPUS --node_rank=0 \
+    --master_port=12346 \
+    autoregressive/sample/sample_t2i_noreverse.py \
+    --yml-path="/home/ma-user/work/z84378256/Selftok-o-inference-siglip2/mimogpt/configs/sd3-siglip2/512-fsq.yml" \
+    --vq-ckpt="/home/ma-user/work/z84378256/model/selftok-tokenizer-vit-block-stage2/iter_123999.pth" \
+    --gpt-model="GPT-XL" \
+    --gpt-ckpt=$local_model_path \
+    --save-image="./results/imagenet_siglip2_noreverse/${parent_dir}-${base_dir}-cfg-${CFG}" \
+    --sd3-pretrained="/home/ma-user/work/selftok/models/sd3_medium.ckpt" \
+    --t5-path="/home/ma-user/work/selftok/models" \
+    --gpt-type="t2i" \
+    --codebook-size=19683 \
+    --image-size=512 \
+    --cfg-scale $CFG \
+	--top-p 1.0 \
+	--top-k 0 \
+	--temperature 1.0 \
+

llamagen-siglip-sb-block-causal/scripts/autoregressive/sample_t2i_parti.sh

@@ -0,0 +1,11 @@
+# !/bin/bash
+set -x
+
+torchrun \
+--nnodes=1 --nproc_per_node=8 --node_rank=0 \
+--master_port=12347 \
+autoregressive/sample/sample_t2i_ddp.py \
+--prompt-csv evaluations/t2i/PartiPrompts.tsv \
+--sample-dir samples_parti \
+--vq-ckpt ./pretrained_models/vq_ds16_t2i.pt \
+"$@"

llamagen-siglip-sb-block-causal/scripts/autoregressive/train_c2i.sh

@@ -0,0 +1,20 @@
+# !/bin/bash
+set -x
+
+export PYTHONPATH="${PWD}:${PYTHONPATH}"
+# export WANDB_NAME="run_$(date +%Y%m%d_%H%M%S)"
+# export WANDB_PROJECT="c2i_selftok"
+# export WANDB_API_KEY="0b30f581d65172381c1f1a45f928210cab80f1de"
+
+
+torchrun --nnodes=1 --nproc_per_node=8 --node_rank=0 \
+    --master_port=20045 \
+    autoregressive/train/train_c2i.py \
+    --code-path /home/ma-user/work/w84402926/SelfTok/llamagen/code-imagenet-E31 \
+    --cloud-save-path ./results_ckpt_e31 \
+    --gpt-model GPT-XL --gpt-type c2i \
+    --vocab-size 16384 \
+    --image-size 512 \
+    --global-batch-size 8 \
+    --no-local-save \
+    --no-wandb

llamagen-siglip-sb-block-causal/scripts/autoregressive/train_c2i_fsdp.sh

@@ -0,0 +1,113 @@
+# 配置CANN相关环境变量
+CANN_INSTALL_PATH_CONF='/etc/Ascend/ascend_cann_install.info'
+sudo cat /etc/Ascend/ascend_cann_install.info
+DEFAULT_CANN_INSTALL_PATH="/usr/local/Ascend/"
+CANN_INSTALL_PATH="/usr/local/Ascend/"
+
+if [ -d ${CANN_INSTALL_PATH}/ascend-toolkit/latest ];then
+  cat ${CANN_INSTALL_PATH}/ascend-toolkit/set_env.sh
+  source ${CANN_INSTALL_PATH}/ascend-toolkit/set_env.sh
+else
+  cat ${CANN_INSTALL_PATH}/nnae/set_env.sh
+  source ${CANN_INSTALL_PATH}/nnae/set_env.sh
+fi
+
+# 导入依赖库
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/local/openblas/lib
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/local/lib/
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/lib64/
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/lib/
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/lib/aarch64_64-linux-gnu
+
+# 配置自定义环境变量
+export HCCL_WHITELIST_DISABLE=1
+
+# log
+export ASCEND_SLOG_PRINT_TO_STDOUT=0   # 日志打屏, 可选
+export ASCEND_GLOBAL_LOG_LEVEL=3       # 日志级别常用 1 INFO级别; 3 ERROR级别
+export ASCEND_GLOBAL_EVENT_ENABLE=0    # 默认不使能event日志信息
+export ASCEND_LAUNCH_BLOCKING=0 # 控制算子执行时是否启动同步模式, 1 同步; 0 异步
+
+# 系统默认环境变量，不建议修改
+MASTER_HOST="$VC_WORKER_HOSTS"
+MASTER_ADDR="${VC_WORKER_HOSTS%%,*}"
+NNODES="$MA_NUM_HOSTS"
+NODE_RANK="$VC_TASK_INDEX"
+NGPUS_PER_NODE="$MA_NUM_GPUS"
+NUM_PROCESSES=$(($NGPUS_PER_NODE * $NNODES))
+
+echo "num_machines: ${NNODES}" >> ${DEEPSPEED_CONFIG_PATH}
+echo "num_processes: ${NUM_PROCESSES}" >> ${DEEPSPEED_CONFIG_PATH}
+
+echo "------> system config <------"
+echo "VC_WORKER_HOSTS: ${VC_WORKER_HOSTS}"
+echo "MASTER_HOST: ${MASTER_HOST}"
+echo "MASTER_ADDR: ${MASTER_ADDR}"
+echo "NNODES: ${NNODES}"
+echo "NODE_RANK: ${NODE_RANK}"
+echo "NGPUS_PER_NODE: ${NGPUS_PER_NODE}"
+echo "NUM_PROCESSES: ${NUM_PROCESSES}"
+echo "${MA_JOB_DIR}"
+echo "------>  <------"
+
+# https://www.hiascend.com/document/detail/zh/canncommercial/63RC2/modeldevpt/ptmigr/ptmigr_0022.html
+export HCCL_WHITELIST_DISABLE=1
+
+if [[ $NODE_RANK == 0 ]]; then
+    EXT_ARGS="--rdzv_conf=is_host=1"
+else
+    EXT_ARGS=""
+fi
+
+# set npu plog env, https://3ms.huawei.com/hi/group/3225441/wiki_6402466.html
+ma_vj_name=`echo ${MA_VJ_NAME} | sed 's:ma-job:modelarts-job:g'`
+task_name="worker-${VC_TASK_INDEX}"
+task_plog_path=${MA_LOG_DIR}/${ma_vj_name}/${task_name}
+
+mkdir -p ${task_plog_path}
+export ASCEND_PROCESS_LOG_PATH=${task_plog_path}
+
+echo "plog path: ${ASCEND_PROCESS_LOG_PATH}"
+
+npu-smi info
+pip install wandb
+
+# set hccl timeout time in seconds
+export HCCL_CONNECT_TIMEOUT=7200
+
+export OMP_NUM_THREADS=4
+export NUMEXPR_MAX_THREADS=4
+
+
+echo "------> pwd <------"
+pwd
+echo "------> files <------"
+ls
+
+export ROOT='/cache'
+python tools/mox_copy.py s3://bucket-2588-wuhu/code/p84402465/code-imagenet-E31/ ${ROOT}/code-imagenet-E31/
+
+
+export MASTER_PORT="${MASTER_PORT:-12335}"
+
+CMD="torchrun \
+  --nnodes=${NNODES} \
+  --nproc_per_node=${NGPUS_PER_NODE} \
+  --node_rank=${NODE_RANK} \
+  --master_addr=${MASTER_ADDR} \
+  --master_port=${MASTER_PORT} \
+  autoregressive/train/train_c2i_fsdp.py \
+  --code-path /cache/code-imagenet-E31 \
+  --cloud-save-path results_ckpt_e31_rope \
+  --train-url s3://bucket-2588-wuhu/outputs/p84402465/llamagen-E31-0811 \
+  --gpt-model GPT-XL \
+  --gpt-type c2i \
+  --vocab-size 16384 \
+  --image-size 512 \
+  --global-batch-size 1024 \
+  --no-local-save \
+  --no-wandb"
+
+echo "------> CMD <------"
+echo "$CMD"
+eval "$CMD"

llamagen-siglip-sb-block-causal/scripts/autoregressive/train_t2i.sh

@@ -0,0 +1,25 @@
+# !/bin/bash
+set -x
+export PYTHONPATH="${PWD}:${PYTHONPATH}"
+export WANDB_NAME="run_$(date +%Y%m%d_%H%M%S)"
+export WANDB_PROJECT="c2i_selftok"
+export WANDB_API_KEY="0b30f581d65172381c1f1a45f928210cab80f1de"
+
+torchrun --nnodes=1 --nproc_per_node=1 --node_rank=0 \
+    --master_port=20045 \
+    autoregressive/train/train_t2i.py \
+    --image-token-path /home/ma-user/work/selftok/imagenet/imagenet_vit_sb_v16_selftok_tokens \
+    --text-token-path /home/ma-user/work/selftok/imagenet/imagenet_cap512_cap_codes \
+    --gpt-model GPT-XL --gpt-type t2i \
+    --dataset t2i_code \
+    --cls-token-num 256 \
+    --vocab-size 16384 \
+    --image-size 512 \
+    --global-batch-size 8 \
+    --ckpt-every 500 \
+    --gradient-accumulation-steps 2 \
+    --cloud-save-path /cache/results_E31_t2i \
+    --s3-save-path s3://bucket-5125-guiyang/selftok/outputs/llamagen/Llamagen-t2i-SigLip \
+    --no-local-save \
+    --no-wandb
+    

llamagen-siglip-sb-block-causal/scripts/autoregressive/train_t2i_multi_node.sh

@@ -0,0 +1,160 @@
+# # !/bin/bash
+# set -x
+# export PYTHONPATH="${PWD}:${PYTHONPATH}"
+# export WANDB_NAME="run_$(date +%Y%m%d_%H%M%S)"
+# export WANDB_PROJECT="c2i_selftok"
+# export WANDB_API_KEY="0b30f581d65172381c1f1a45f928210cab80f1de"
+
+# torchrun --nnodes=1 --nproc_per_node=8 --node_rank=0 \
+#     --master_port=20045 \
+#     autoregressive/train/train_t2i.py \
+#     --image-token-path /cache/code-imagenet-E31_cap/imagenet_cap512_image_codes \
+#     --text-token-path /cache/code-imagenet-E31_cap/imagenet_cap512_cap_codes \
+#     --gpt-model GPT-XL --gpt-type t2i \
+#     --dataset t2i_code \
+#     --cls-token-num 256 \
+#     --vocab-size 16384 \
+#     --image-size 512 \
+#     --global-batch-size 512 \
+#     --gradient-accumulation-steps 2 \
+#     --cloud-save-path /cache/results_E31_t2i \
+#     --no-local-save \
+#     --no-wandb
+
+
+
+
+
+    # 配置CANN相关环境变量
+CANN_INSTALL_PATH_CONF='/etc/Ascend/ascend_cann_install.info'
+sudo cat /etc/Ascend/ascend_cann_install.info
+DEFAULT_CANN_INSTALL_PATH="/usr/local/Ascend/"
+CANN_INSTALL_PATH="/usr/local/Ascend/"
+
+if [ -d ${CANN_INSTALL_PATH}/ascend-toolkit/latest ];then
+  cat ${CANN_INSTALL_PATH}/ascend-toolkit/set_env.sh
+  source ${CANN_INSTALL_PATH}/ascend-toolkit/set_env.sh
+else
+  cat ${CANN_INSTALL_PATH}/nnae/set_env.sh
+  source ${CANN_INSTALL_PATH}/nnae/set_env.sh
+fi
+
+# 导入依赖库
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/local/openblas/lib
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/local/lib/
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/lib64/
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/lib/
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/lib/aarch64_64-linux-gnu
+
+# 配置自定义环境变量
+export HCCL_WHITELIST_DISABLE=1
+
+# log
+export ASCEND_SLOG_PRINT_TO_STDOUT=0   # 日志打屏, 可选
+export ASCEND_GLOBAL_LOG_LEVEL=3       # 日志级别常用 1 INFO级别; 3 ERROR级别
+export ASCEND_GLOBAL_EVENT_ENABLE=0    # 默认不使能event日志信息
+export ASCEND_LAUNCH_BLOCKING=0 # 控制算子执行时是否启动同步模式, 1 同步; 0 异步
+
+# 系统默认环境变量，不建议修改
+MASTER_HOST="$VC_WORKER_HOSTS"
+MASTER_ADDR="${VC_WORKER_HOSTS%%,*}"
+NNODES="$MA_NUM_HOSTS"
+NODE_RANK="$VC_TASK_INDEX"
+NGPUS_PER_NODE="$MA_NUM_GPUS"
+NUM_PROCESSES=$(($NGPUS_PER_NODE * $NNODES))
+
+echo "num_machines: ${NNODES}" >> ${DEEPSPEED_CONFIG_PATH}
+echo "num_processes: ${NUM_PROCESSES}" >> ${DEEPSPEED_CONFIG_PATH}
+
+echo "------> system config <------"
+echo "VC_WORKER_HOSTS: ${VC_WORKER_HOSTS}"
+echo "MASTER_HOST: ${MASTER_HOST}"
+echo "MASTER_ADDR: ${MASTER_ADDR}"
+echo "NNODES: ${NNODES}"
+echo "NODE_RANK: ${NODE_RANK}"
+echo "NGPUS_PER_NODE: ${NGPUS_PER_NODE}"
+echo "NUM_PROCESSES: ${NUM_PROCESSES}"
+echo "${MA_JOB_DIR}"
+echo "------>  <------"
+
+# https://www.hiascend.com/document/detail/zh/canncommercial/63RC2/modeldevpt/ptmigr/ptmigr_0022.html
+export HCCL_WHITELIST_DISABLE=1
+
+if [[ $NODE_RANK == 0 ]]; then
+    EXT_ARGS="--rdzv_conf=is_host=1"
+else
+    EXT_ARGS=""
+fi
+
+# set npu plog env, https://3ms.huawei.com/hi/group/3225441/wiki_6402466.html
+ma_vj_name=`echo ${MA_VJ_NAME} | sed 's:ma-job:modelarts-job:g'`
+task_name="worker-${VC_TASK_INDEX}"
+task_plog_path=${MA_LOG_DIR}/${ma_vj_name}/${task_name}
+
+mkdir -p ${task_plog_path}
+export ASCEND_PROCESS_LOG_PATH=${task_plog_path}
+
+echo "plog path: ${ASCEND_PROCESS_LOG_PATH}"
+
+npu-smi info
+pip install wandb
+
+# set hccl timeout time in seconds
+export HCCL_CONNECT_TIMEOUT=7200
+
+export OMP_NUM_THREADS=4
+export NUMEXPR_MAX_THREADS=4
+
+
+echo "------> pwd <------"
+pwd
+echo "------> files <------"
+ls
+
+export ROOT='/cache'
+# python tools/mox_copy.py s3://bucket-2588-wuhu/code/p84402465/code-imagenet-E31/ ${ROOT}/code-imagenet-E31/
+
+
+export MASTER_PORT="${MASTER_PORT:-12335}"
+
+CMD="torchrun \
+  --nnodes=${NNODES} \
+  --nproc_per_node=${NGPUS_PER_NODE} \
+  --node_rank=${NODE_RANK} \
+  --master_addr=${MASTER_ADDR} \
+  --master_port=${MASTER_PORT} \
+  autoregressive/train/train_t2i.py \
+    --image-token-path  /home/ma-user/modelarts/user-job-dir/sfs/selftok/imagenet/imagenet_cap512_cap_codes/imagenet_cap512_image_codes \
+    --text-token-path /home/ma-user/modelarts/user-job-dir/sfs/selftok/imagenet/code-imagenet-E31_cap/imagenet_cap512_cap_codes \
+    --gpt-model GPT-XL --gpt-type t2i \
+    --dataset t2i_code \
+    --cls-token-num 256 \
+    --vocab-size 16384 \
+    --image-size 512 \
+    --global-batch-size 512 \
+    --ckpt-every 5000 \
+    --gradient-accumulation-steps 2 \
+    --cloud-save-path /cache/results_Selftok-Llamagen-t2i-RelPosEmb \
+    --s3-save-path s3://bucket-2588-wuhu/code/w84402926/experiment/llamagen_selftok/Selftok-Llamagen-t2i-RelPosEmb \
+    --no-local-save \
+    --no-wandb"
+
+# torchrun --nnodes=1 --nproc_per_node=8 --node_rank=0 \
+#     --master_port=20045 \
+#     autoregressive/train/train_t2i.py \
+#     --image-token-path /cache/code-imagenet-E31_cap/imagenet_cap512_image_codes \
+#     --text-token-path /cache/code-imagenet-E31_cap/imagenet_cap512_cap_codes \
+#     --gpt-model GPT-XL --gpt-type t2i \
+#     --dataset t2i_code \
+#     --cls-token-num 256 \
+#     --vocab-size 16384 \
+#     --image-size 512 \
+#     --global-batch-size 512 \
+#     --gradient-accumulation-steps 2 \
+#     --cloud-save-path /cache/results_E31_t2i \
+#     --no-local-save \
+#     --no-wandb
+
+echo "------> CMD <------"
+echo "$CMD"
+eval "$CMD"

llamagen-siglip-sb-block-causal/scripts/autoregressive/train_t2i_multi_node_absPos.sh

@@ -0,0 +1,164 @@
+# # !/bin/bash
+# set -x
+# export PYTHONPATH="${PWD}:${PYTHONPATH}"
+# export WANDB_NAME="run_$(date +%Y%m%d_%H%M%S)"
+# export WANDB_PROJECT="c2i_selftok"
+# export WANDB_API_KEY="0b30f581d65172381c1f1a45f928210cab80f1de"
+
+# torchrun --nnodes=1 --nproc_per_node=8 --node_rank=0 \
+#     --master_port=20045 \
+#     autoregressive/train/train_t2i.py \
+#     --image-token-path /cache/code-imagenet-E31_cap/imagenet_cap512_image_codes \
+#     --text-token-path /cache/code-imagenet-E31_cap/imagenet_cap512_cap_codes \
+#     --gpt-model GPT-XL --gpt-type t2i \
+#     --dataset t2i_code \
+#     --cls-token-num 256 \
+#     --vocab-size 16384 \
+#     --image-size 512 \
+#     --global-batch-size 512 \
+#     --gradient-accumulation-steps 2 \
+#     --cloud-save-path /cache/results_E31_t2i \
+#     --no-local-save \
+#     --no-wandb
+
+
+
+
+
+    # 配置CANN相关环境变量
+CANN_INSTALL_PATH_CONF='/etc/Ascend/ascend_cann_install.info'
+sudo cat /etc/Ascend/ascend_cann_install.info
+DEFAULT_CANN_INSTALL_PATH="/usr/local/Ascend/"
+CANN_INSTALL_PATH="/usr/local/Ascend/"
+
+if [ -d ${CANN_INSTALL_PATH}/ascend-toolkit/latest ];then
+  cat ${CANN_INSTALL_PATH}/ascend-toolkit/set_env.sh
+  source ${CANN_INSTALL_PATH}/ascend-toolkit/set_env.sh
+else
+  cat ${CANN_INSTALL_PATH}/nnae/set_env.sh
+  source ${CANN_INSTALL_PATH}/nnae/set_env.sh
+fi
+
+# 导入依赖库
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/local/openblas/lib
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/local/lib/
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/lib64/
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/lib/
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/lib/aarch64_64-linux-gnu
+
+# 配置自定义环境变量
+export HCCL_WHITELIST_DISABLE=1
+
+# log
+export ASCEND_SLOG_PRINT_TO_STDOUT=0   # 日志打屏, 可选
+export ASCEND_GLOBAL_LOG_LEVEL=3       # 日志级别常用 1 INFO级别; 3 ERROR级别
+export ASCEND_GLOBAL_EVENT_ENABLE=0    # 默认不使能event日志信息
+export ASCEND_LAUNCH_BLOCKING=0 # 控制算子执行时是否启动同步模式, 1 同步; 0 异步
+
+# 系统默认环境变量，不建议修改
+MASTER_HOST="$VC_WORKER_HOSTS"
+MASTER_ADDR="${VC_WORKER_HOSTS%%,*}"
+NNODES="$MA_NUM_HOSTS"
+NODE_RANK="$VC_TASK_INDEX"
+NGPUS_PER_NODE="$MA_NUM_GPUS"
+NUM_PROCESSES=$(($NGPUS_PER_NODE * $NNODES))
+
+echo "num_machines: ${NNODES}" >> ${DEEPSPEED_CONFIG_PATH}
+echo "num_processes: ${NUM_PROCESSES}" >> ${DEEPSPEED_CONFIG_PATH}
+
+echo "------> system config <------"
+echo "VC_WORKER_HOSTS: ${VC_WORKER_HOSTS}"
+echo "MASTER_HOST: ${MASTER_HOST}"
+echo "MASTER_ADDR: ${MASTER_ADDR}"
+echo "NNODES: ${NNODES}"
+echo "NODE_RANK: ${NODE_RANK}"
+echo "NGPUS_PER_NODE: ${NGPUS_PER_NODE}"
+echo "NUM_PROCESSES: ${NUM_PROCESSES}"
+echo "${MA_JOB_DIR}"
+echo "------>  <------"
+
+# https://www.hiascend.com/document/detail/zh/canncommercial/63RC2/modeldevpt/ptmigr/ptmigr_0022.html
+export HCCL_WHITELIST_DISABLE=1
+
+if [[ $NODE_RANK == 0 ]]; then
+    EXT_ARGS="--rdzv_conf=is_host=1"
+else
+    EXT_ARGS=""
+fi
+
+# set npu plog env, https://3ms.huawei.com/hi/group/3225441/wiki_6402466.html
+ma_vj_name=`echo ${MA_VJ_NAME} | sed 's:ma-job:modelarts-job:g'`
+task_name="worker-${VC_TASK_INDEX}"
+task_plog_path=${MA_LOG_DIR}/${ma_vj_name}/${task_name}
+
+mkdir -p ${task_plog_path}
+export ASCEND_PROCESS_LOG_PATH=${task_plog_path}
+
+echo "plog path: ${ASCEND_PROCESS_LOG_PATH}"
+
+npu-smi info
+pip install wandb
+pip install einx
+pip install fairscale
+pip install lpips
+pip install diffusers==0.31.0
+
+# set hccl timeout time in seconds
+export HCCL_CONNECT_TIMEOUT=7200
+
+export OMP_NUM_THREADS=4
+export NUMEXPR_MAX_THREADS=4
+
+
+echo "------> pwd <------"
+pwd
+echo "------> files <------"
+ls
+
+export ROOT='/cache'
+# python tools/mox_copy.py s3://bucket-2588-wuhu/code/p84402465/code-imagenet-E31/ ${ROOT}/code-imagenet-E31/
+
+
+export MASTER_PORT="${MASTER_PORT:-12335}"
+
+CMD="torchrun \
+  --nnodes=${NNODES} \
+  --nproc_per_node=${NGPUS_PER_NODE} \
+  --node_rank=${NODE_RANK} \
+  --master_addr=${MASTER_ADDR} \
+  --master_port=${MASTER_PORT} \
+  autoregressive/train/train_t2i.py \
+    --image-token-path  /home/ma-user/modelarts/user-job-dir/sfs/selftok/imagenet/imagenet_vit_sb_v16_selftok_tokens \
+    --text-token-path /home/ma-user/modelarts/user-job-dir/sfs/selftok/imagenet/imagenet_cap512_cap_codes \
+    --gpt-model GPT-XL --gpt-type t2i \
+    --dataset t2i_code \
+    --cls-token-num 256 \
+    --vocab-size 19683 \
+    --image-size 512 \
+    --global-batch-size 256 \
+    --ckpt-every 5000 \
+    --gradient-accumulation-steps 2 \
+    --cloud-save-path /cache/results_Selftok-Llamagen-t2i-AbsPosEmb \
+    --s3-save-path s3://bucket-5125-guiyang/selftok/outputs/llamagen_selftok/Selftok-Llamagen-t2i-vit-sb-v16-blockcausal \
+    --no-local-save \
+    --no-wandb"
+
+# torchrun --nnodes=1 --nproc_per_node=8 --node_rank=0 \
+#     --master_port=20045 \
+#     autoregressive/train/train_t2i.py \
+#     --image-token-path /cache/code-imagenet-E31_cap/imagenet_cap512_image_codes \
+#     --text-token-path /cache/code-imagenet-E31_cap/imagenet_cap512_cap_codes \
+#     --gpt-model GPT-XL --gpt-type t2i \
+#     --dataset t2i_code \
+#     --cls-token-num 256 \
+#     --vocab-size 16384 \
+#     --image-size 512 \
+#     --global-batch-size 512 \
+#     --gradient-accumulation-steps 2 \
+#     --cloud-save-path /cache/results_E31_t2i \
+#     --no-local-save \
+#     --no-wandb
+
+echo "------> CMD <------"
+echo "$CMD"
+eval "$CMD"

llamagen-siglip-sb-block-causal/scripts/autoregressive/train_t2i_multi_node_overfit.sh

@@ -0,0 +1,164 @@
+# # !/bin/bash
+# set -x
+# export PYTHONPATH="${PWD}:${PYTHONPATH}"
+# export WANDB_NAME="run_$(date +%Y%m%d_%H%M%S)"
+# export WANDB_PROJECT="c2i_selftok"
+# export WANDB_API_KEY="0b30f581d65172381c1f1a45f928210cab80f1de"
+
+# torchrun --nnodes=1 --nproc_per_node=8 --node_rank=0 \
+#     --master_port=20045 \
+#     autoregressive/train/train_t2i.py \
+#     --image-token-path /cache/code-imagenet-E31_cap/imagenet_cap512_image_codes \
+#     --text-token-path /cache/code-imagenet-E31_cap/imagenet_cap512_cap_codes \
+#     --gpt-model GPT-XL --gpt-type t2i \
+#     --dataset t2i_code \
+#     --cls-token-num 256 \
+#     --vocab-size 16384 \
+#     --image-size 512 \
+#     --global-batch-size 512 \
+#     --gradient-accumulation-steps 2 \
+#     --cloud-save-path /cache/results_E31_t2i \
+#     --no-local-save \
+#     --no-wandb
+
+
+
+
+
+    # 配置CANN相关环境变量
+CANN_INSTALL_PATH_CONF='/etc/Ascend/ascend_cann_install.info'
+sudo cat /etc/Ascend/ascend_cann_install.info
+DEFAULT_CANN_INSTALL_PATH="/usr/local/Ascend/"
+CANN_INSTALL_PATH="/usr/local/Ascend/"
+
+if [ -d ${CANN_INSTALL_PATH}/ascend-toolkit/latest ];then
+  cat ${CANN_INSTALL_PATH}/ascend-toolkit/set_env.sh
+  source ${CANN_INSTALL_PATH}/ascend-toolkit/set_env.sh
+else
+  cat ${CANN_INSTALL_PATH}/nnae/set_env.sh
+  source ${CANN_INSTALL_PATH}/nnae/set_env.sh
+fi
+
+# 导入依赖库
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/local/openblas/lib
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/local/lib/
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/lib64/
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/lib/
+export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:/usr/lib/aarch64_64-linux-gnu
+
+# 配置自定义环境变量
+export HCCL_WHITELIST_DISABLE=1
+
+# log
+export ASCEND_SLOG_PRINT_TO_STDOUT=0   # 日志打屏, 可选
+export ASCEND_GLOBAL_LOG_LEVEL=3       # 日志级别常用 1 INFO级别; 3 ERROR级别
+export ASCEND_GLOBAL_EVENT_ENABLE=0    # 默认不使能event日志信息
+export ASCEND_LAUNCH_BLOCKING=0 # 控制算子执行时是否启动同步模式, 1 同步; 0 异步
+
+# 系统默认环境变量，不建议修改
+MASTER_HOST="$VC_WORKER_HOSTS"
+MASTER_ADDR="${VC_WORKER_HOSTS%%,*}"
+NNODES="$MA_NUM_HOSTS"
+NODE_RANK="$VC_TASK_INDEX"
+NGPUS_PER_NODE="$MA_NUM_GPUS"
+NUM_PROCESSES=$(($NGPUS_PER_NODE * $NNODES))
+
+echo "num_machines: ${NNODES}" >> ${DEEPSPEED_CONFIG_PATH}
+echo "num_processes: ${NUM_PROCESSES}" >> ${DEEPSPEED_CONFIG_PATH}
+
+echo "------> system config <------"
+echo "VC_WORKER_HOSTS: ${VC_WORKER_HOSTS}"
+echo "MASTER_HOST: ${MASTER_HOST}"
+echo "MASTER_ADDR: ${MASTER_ADDR}"
+echo "NNODES: ${NNODES}"
+echo "NODE_RANK: ${NODE_RANK}"
+echo "NGPUS_PER_NODE: ${NGPUS_PER_NODE}"
+echo "NUM_PROCESSES: ${NUM_PROCESSES}"
+echo "${MA_JOB_DIR}"
+echo "------>  <------"
+
+# https://www.hiascend.com/document/detail/zh/canncommercial/63RC2/modeldevpt/ptmigr/ptmigr_0022.html
+export HCCL_WHITELIST_DISABLE=1
+
+if [[ $NODE_RANK == 0 ]]; then
+    EXT_ARGS="--rdzv_conf=is_host=1"
+else
+    EXT_ARGS=""
+fi
+
+# set npu plog env, https://3ms.huawei.com/hi/group/3225441/wiki_6402466.html
+ma_vj_name=`echo ${MA_VJ_NAME} | sed 's:ma-job:modelarts-job:g'`
+task_name="worker-${VC_TASK_INDEX}"
+task_plog_path=${MA_LOG_DIR}/${ma_vj_name}/${task_name}
+
+mkdir -p ${task_plog_path}
+export ASCEND_PROCESS_LOG_PATH=${task_plog_path}
+
+echo "plog path: ${ASCEND_PROCESS_LOG_PATH}"
+
+npu-smi info
+pip install wandb
+pip install einx
+pip install fairscale
+pip install lpips
+pip install diffusers==0.31.0
+
+# set hccl timeout time in seconds
+export HCCL_CONNECT_TIMEOUT=7200
+
+export OMP_NUM_THREADS=4
+export NUMEXPR_MAX_THREADS=4
+
+
+echo "------> pwd <------"
+pwd
+echo "------> files <------"
+ls
+
+export ROOT='/cache'
+# python tools/mox_copy.py s3://bucket-2588-wuhu/code/p84402465/code-imagenet-E31/ ${ROOT}/code-imagenet-E31/
+
+
+export MASTER_PORT="${MASTER_PORT:-12335}"
+
+CMD="torchrun \
+  --nnodes=${NNODES} \
+  --nproc_per_node=${NGPUS_PER_NODE} \
+  --node_rank=${NODE_RANK} \
+  --master_addr=${MASTER_ADDR} \
+  --master_port=${MASTER_PORT} \
+  autoregressive/train/train_t2i.py \
+    --image-token-path  /home/ma-user/modelarts/user-job-dir/sfs/selftok/imagenet/imagenet_vit_sb_v16_selftok_tokens \
+    --text-token-path /home/ma-user/modelarts/user-job-dir/sfs/selftok/imagenet/overfits \
+    --gpt-model GPT-XL --gpt-type t2i \
+    --dataset t2i_code \
+    --cls-token-num 256 \
+    --vocab-size 19683 \
+    --image-size 512 \
+    --global-batch-size 256 \
+    --ckpt-every 5000 \
+    --gradient-accumulation-steps 2 \
+    --cloud-save-path /cache/results_Selftok-Llamagen-t2i-AbsPosEmb \
+    --s3-save-path s3://bucket-5125-guiyang/selftok/outputs/llamagen_selftok/Selftok-Llamagen-t2i-vit-sb-v16-blockcausal-overfit \
+    --no-local-save \
+    --no-wandb"
+
+# torchrun --nnodes=1 --nproc_per_node=8 --node_rank=0 \
+#     --master_port=20045 \
+#     autoregressive/train/train_t2i.py \
+#     --image-token-path /cache/code-imagenet-E31_cap/imagenet_cap512_image_codes \
+#     --text-token-path /cache/code-imagenet-E31_cap/imagenet_cap512_cap_codes \
+#     --gpt-model GPT-XL --gpt-type t2i \
+#     --dataset t2i_code \
+#     --cls-token-num 256 \
+#     --vocab-size 16384 \
+#     --image-size 512 \
+#     --global-batch-size 512 \
+#     --gradient-accumulation-steps 2 \
+#     --cloud-save-path /cache/results_E31_t2i \
+#     --no-local-save \
+#     --no-wandb
+
+echo "------> CMD <------"
+echo "$CMD"
+eval "$CMD"

llamagen-siglip-sb-block-causal/scripts/autoregressive/train_t2i_stage1.sh

@@ -0,0 +1,13 @@
+# !/bin/bash
+set -x
+
+torchrun \
+--nnodes=$nnodes --nproc_per_node=$nproc_per_node --node_rank=$node_rank \
+--master_addr=$master_addr --master_port=$master_port \
+autoregressive/train/train_t2i.py \
+--vq-ckpt ./pretrained_models/vq_ds16_t2i.pt \
+--data-path /path/to/laion_coco50M \
+--t5-feat-path /path/to/laion_coco50M_flan_t5_xl \
+--dataset t2i \
+--image-size 256 \
+"$@"

llamagen-siglip-sb-block-causal/scripts/autoregressive/train_t2i_stage2.sh

@@ -0,0 +1,14 @@
+# !/bin/bash
+set -x
+
+torchrun \
+--nnodes=$nnodes --nproc_per_node=$nproc_per_node --node_rank=$node_rank \
+--master_addr=$master_addr --master_port=$master_port \
+autoregressive/train/train_t2i.py \
+--vq-ckpt ./pretrained_models/vq_ds16_t2i.pt \
+--data-path /path/to/high_aesthetic_10M \
+--t5-feat-path /path/to/high_aesthetic_10M_flan_t5_xl \
+--short-t5-feat-path /path/to/high_aesthetic_10M_trunc_flan_t5_xl \
+--dataset t2i \
+--image-size 512 \
+"$@"

llamagen-siglip-sb-block-causal/scripts/language/extract_flan_t5_feat_laion_coco_stage1.sh

@@ -0,0 +1,11 @@
+# !/bin/bash
+set -x
+
+torchrun \
+--nnodes=1 --nproc_per_node=8 --node_rank=0 \
+--master_port=12337 \
+language/extract_t5_feature.py \
+--data-path /path/to/laion_coco50M \
+--t5-path /path/to/laion_coco50M_flan_t5_xl \
+--caption-key blip \
+"$@"

llamagen-siglip-sb-block-causal/scripts/language/extract_flan_t5_feat_stage2.sh

@@ -0,0 +1,10 @@
+# !/bin/bash
+set -x
+
+torchrun \
+--nnodes=1 --nproc_per_node=8 --node_rank=0 \
+--master_port=12337 \
+language/extract_t5_feature.py \
+--data-path /path/to/high_aesthetic_10M \
+--t5-path /path/to/high_aesthetic_10M_flan_t5_xl \
+"$@"

llamagen-siglip-sb-block-causal/scripts/language/extract_flan_t5_feat_trunc_stage2.sh

@@ -0,0 +1,11 @@
+# !/bin/bash
+set -x
+
+torchrun \
+--nnodes=1 --nproc_per_node=8 --node_rank=0 \
+--master_port=12337 \
+language/extract_t5_feature.py \
+--data-path /path/to/high_aesthetic_10M \
+--t5-path /path/to/high_aesthetic_10M_trunc_flan_t5_xl \
+--trunc-caption \
+"$@"

llamagen-siglip-sb-block-causal/scripts/language/extract_flan_t5_imagenet.sh

@@ -0,0 +1,9 @@
+# 单机多卡（建议用 torchrun 或你现有的 init_distributed_mode）
+torchrun --nproc_per_node=8 language/extract_t5_feature_imagenet.py \
+  --parquet-dir /home/ma-user/work/w84402926/SelfTok/llamagen/imagenet/imagenet-1k-vl-enriched \
+  --out-dir /cache/code-imagenet-E31_cap \
+  --data-start 0 --data-end 206 \
+  --pattern "train-{idx:05d}-of-00207.parquet" \
+  --t5-model-path /home/ma-user/work/w84402926/SelfTok/llamagen/imagenet \
+  --t5-model-type flan-t5-xl \
+  --precision bf16

llamagen-siglip-sb-block-causal/scripts/tokenizer/reconstruction_consistency_decoder.sh

@@ -0,0 +1,8 @@
+# !/bin/bash
+set -x
+
+torchrun \
+--nnodes=1 --nproc_per_node=8 --node_rank=0 \
+--master_port=12344 \
+tokenizer/consistencydecoder/reconstruction_cd_ddp.py \
+"$@"

llamagen-siglip-sb-block-causal/scripts/tokenizer/reconstruction_vae.sh

@@ -0,0 +1,8 @@
+# !/bin/bash
+set -x
+
+torchrun \
+--nnodes=1 --nproc_per_node=8 --node_rank=0 \
+--master_port=12344 \
+tokenizer/vae/reconstruction_vae_ddp.py \
+"$@"

llamagen-siglip-sb-block-causal/scripts/tokenizer/reconstruction_vq.sh

@@ -0,0 +1,8 @@
+# !/bin/bash
+set -x
+
+torchrun \
+--nnodes=1 --nproc_per_node=8 --node_rank=0 \
+--master_port=12344 \
+tokenizer/tokenizer_image/reconstruction_vq_ddp.py \
+"$@"

llamagen-siglip-sb-block-causal/scripts/tokenizer/reconstruction_vqgan.sh

@@ -0,0 +1,8 @@
+# !/bin/bash
+set -x
+
+torchrun \
+--nnodes=1 --nproc_per_node=8 --node_rank=0 \
+--master_port=12344 \
+tokenizer/vqgan/reconstruction_vqgan_ddp.py \
+"$@"

llamagen-siglip-sb-block-causal/scripts/tokenizer/train_vq.sh

@@ -0,0 +1,7 @@
+# !/bin/bash
+set -x
+
+torchrun \
+--nnodes=$nnodes --nproc_per_node=$nproc_per_node --node_rank=$node_rank \
+--master_addr=$master_addr --master_port=$master_port \
+tokenizer/tokenizer_image/vq_train.py "$@"

llamagen-siglip-sb-block-causal/scripts/tokenizer/train_vq_finetune.sh

@@ -0,0 +1,15 @@
+# !/bin/bash
+set -x
+
+torchrun \
+--nnodes=$nnodes --nproc_per_node=$nproc_per_node --node_rank=$node_rank \
+--master_addr=$master_addr --master_port=$master_port \
+tokenizer/tokenizer_image/vq_train.py \
+--finetune \
+--disc-start 0 \
+--vq-ckpt ./pretrained_models/vq_ds16_c2i.pt \
+--dataset t2i_image \
+--data-path /path/to/high_aesthetic_10M \
+--data-face-path /path/to/face_2M \
+--cloud-save-path /path/to/cloud_disk \
+"$@"

llamagen-siglip-sb-block-causal/scripts/tokenizer/train_vq_finetune_continue.sh

@@ -0,0 +1,15 @@
+# !/bin/bash
+set -x
+
+torchrun \
+--nnodes=$nnodes --nproc_per_node=$nproc_per_node --node_rank=$node_rank \
+--master_addr=$master_addr --master_port=$master_port \
+tokenizer/tokenizer_image/vq_train.py \
+--disc-start 0 \
+--dataset t2i_image \
+--data-path /path/to/high_aesthetic_10M \
+--data-face-path /path/to/face_2M \
+--cloud-save-path /path/to/cloud_disk \
+"$@"
+
+# --vq-ckpt xxx.pt

llamagen-siglip-sb-block-causal/scripts/tokenizer/val.sh

@@ -0,0 +1,8 @@
+# !/bin/bash
+set -x
+
+torchrun \
+--nnodes=1 --nproc_per_node=8 --node_rank=0 \
+--master_port=12343 \
+tokenizer/validation/val_ddp.py \
+"$@"